Problem in datastore for training using yolo v2

i created the ground truth table from my own dataset. I follow this tutorial https://www.mathworks.com/help/vision/ref/objectdetectortrainingdata.html?searchHighlight=objectDetectorTrainingData&s_tid=doc_srchtitle; however, in stead of using ACF, i intend to use yolo v2 as the following tutorial https://www.mathworks.com/help/deeplearning/examples/object-detection-using-yolo-v2.html. The code is below.

clc
tic
[imds,bxds] = objectDetectorTrainingData(gTruth);
toc
tic
trainingData = combine(imds,bxds);
toc
data = read(trainingData);
I = data{1};
bbox = data{2};
annotatedImage = insertShape(I,'Rectangle',bbox);
annotatedImage = imresize(annotatedImage,2);
figure
imshow(annotatedImage)
display('create YOLO network');
inputSize = [224 224 3];
numClasses = 100;
trainingDataForEstimation = transform(trainingData,@(data)preprocessData(data,inputSize));
numAnchors = 5;
[anchorBoxes, meanIoU] = estimateAnchorBoxes(trainingDataForEstimation, numAnchors);
display('featureExtraction');
featureExtractionNetwork = resnet50;

featureLayer = 'activation_40_relu';

lgraph = yolov2Layers(inputSize,numClasses,anchorBoxes,featureExtractionNetwork,featureLayer);
display('Data augmentation');

augmentedTrainingData = transform(trainingData,@augmentData);
% Visualize the augmented images.
augmentedData = cell(4,1);
for k = 1:4
    data = read(augmentedTrainingData);
    augmentedData{k} = insertShape(data{1},'Rectangle',data{2});
    reset(augmentedTrainingData);
end

figure
montage(augmentedData,'BorderSize',10)
display('preprocessing data');
preprocessedTrainingData = transform(augmentedTrainingData,@(data)preprocessData(data,inputSize));
data = read(preprocessedTrainingData);
I = data{1};
bbox = data{2};
annotatedImage = insertShape(I,'Rectangle',bbox);
annotatedImage = imresize(annotatedImage,2);

figure
imshow(annotatedImage)
display('train yolo network');
options = trainingOptions('sgdm', ...
       'InitialLearnRate', 0.001, ...
       'Verbose',true, ...
       'MiniBatchSize',5, ...
       'MaxEpochs',10, ...
       'Shuffle','every-epoch', ...
       'VerboseFrequency',10);        
    % Train the YOLO v2 detector.
[detector,info] = trainYOLOv2ObjectDetector(preprocessedTrainingData,lgraph,options);

After a couple of minutes, i receive this error.

Error using vision.internal.cnn.validation.checkTrainingBoxes (line 12)
Training data from a read of the input datastore contains invalid bounding boxes. Bounding boxes must be
non-empty, fully contained within their associated image and must have positive width and height. Use datastore
transform method and remove invalid bounding boxes.

Error in vision.internal.cnn.yolo.validateImagesAndBoxesTransform (line 22)
vision.internal.cnn.validation.checkTrainingBoxes(images, boxes);

Error in
trainYOLOv2ObjectDetector>@(data)vision.internal.cnn.yolo.validateImagesAndBoxesTransform(data,params.InputSize)
(line 262)
        @(data)vision.internal.cnn.yolo.validateImagesAndBoxesTransform(data,params.InputSize));

Error in matlab.io.datastore.TransformedDatastore/read (line 148)
                        data = self.Transforms{ii}(data);

Error in nnet.internal.cnn.DataLoader/manageReadQueue (line 161)
                data = read(self.Datastore);

Error in nnet.internal.cnn.DataLoader/nextBatch (line 88)
                manageReadQueue(self);

Error in nnet.internal.cnn.GeneralDatastoreDispatcher/readNextBatchIntoCache (line 350)
                data = nextBatch(this.DataLoader);

Error in nnet.internal.cnn.GeneralDatastoreDispatcher/next (line 307)
            this.readNextBatchIntoCache();

Error in nnet.internal.cnn.Trainer/doComputeStatistics (line 213)
                X = data.next();

Error in nnet.internal.cnn.Trainer/computeInputStatisticsInEnvironment (line 205)
            stats = doComputeStatistics(this, data, stats, net);

Error in nnet.internal.cnn.Trainer/initializeNetwork (line 70)
                stats = this.computeInputStatisticsInEnvironment(data,stats,net);

Error in vision.internal.cnn.trainNetwork (line 82)
    trainedNet = trainer.initializeNetwork(trainedNet, inputStatsDispatcher);

Error in trainYOLOv2ObjectDetector>iTrainYOLOv2 (line 391)
[yolov2Net, info] = vision.internal.cnn.trainNetwork(...

Error in trainYOLOv2ObjectDetector (line 187)
[net, info] = iTrainYOLOv2(ds, lgraph, params, mapping, options, checkpointSaver);

Error in test_yolo_food_detection (line 61)
    [detector,info] = trainYOLOv2ObjectDetector(preprocessedTrainingData,lgraph,options);

My ground truth table is some kind like that.

    gTruth = 

  groundTruth with properties:

          DataSource: [1×1 groundTruthDataSource]
    LabelDefinitions: [100×4 table]
           LabelData: [8768×100 table]
    
    gTruth.LabelData
    
    rice         eelsonrice        pilaf        chickenAndEggonrice    porkcutletonrice
    ____________    ____________    ____________    ___________________    ________________

    {1×4 double}    {0×0 double}    {0×0 double}       {0×0 double}          {0×0 double}  
    {1×4 double}    {0×0 double}    {0×0 double}       {0×0 double}          {0×0 double}  
    {1×4 double}    {0×0 double}    {0×0 double}       {0×0 double}          {0×0 double}  
    {1×4 double}    {0×0 double}    {0×0 double}       {0×0 double}          {0×0 double}  
    {1×4 double}    {0×0 double}    {0×0 double}       {0×0 double}          {0×0 double}  

There are some images in the datasource do not have the presence of some category; therefore, i assign their lableData is empty.

Does this make the error above? How to fix it?

Thank you.

[Zheng Yuan]

hallo, i am faced with the same problem now when training my YOLOv2 model.

Have you already solved it?

Thank you

Answer 1

[ahmed shahin]

Dear AnH

if yoy please solve this error , let's me know how you did that?

Thanks

Ahmed

Answer 2

[Vivek Akkala]

Having empty bounding boxes in the datasource will cause the error. As mentioned in the error message you can either remove the empty bounding boxes manually or use a datastore transform method to remove these empty bounding boxes.